CloneVI: A deep generative model for deciphering the relationships between cell state and mitochondrial genotype

CloneVI infers clonal structure from a latent cell state and reconstructs mutation counts, recovering ground-truth clonal structure in simulated single-clone and mixed-clone populations.

• Why it matters: Clones are often inferred from mitochondrial mutations alone, even though cell states can be non-genetically regulated; linking state to mitochondrial genotype is key for understanding clone-driven disease.
• What they did: The model learns a latent cell state from gene expression or chromatin accessibility, then estimates per-cell clonal probabilities and clone allele frequencies by reconstructing observed mutation counts.
• The result: In chronic lymphocytic leukemia mitochondrial single-cell ATAC-seq, it separated populations with different heteroplasmy levels and identified increased PAX9 accessibility, enabling mutation-defined populations tied to cell states.

AI summary of the abstract · may contain errors · not medical advice · read the paper.